Medicaid Work Requirements

From January 2027, adults in the Medicaid expansion group must document 80 hours a month of work, study or volunteering to keep their coverage.

AI evaluation · not yet reviewed by a human

This evaluation was produced and sourced by an AI model; a human review is still pending. Figures and conclusions may still change. The review log is at the foot of the page.How review works →

The 2025 reconciliation law requires every state to check, before enrolling an adult aged 19 to 64 in the expansion group and at least twice a year afterwards, whether that person worked, studied or volunteered for at least 80 hours in the preceding month. People who cannot document it lose their coverage; pregnancy, disability, and care of a young child are among the exemptions. States must use data they already hold before asking the enrollee, and the federal government pays part of the cost of building the systems that do it. Nebraska began in May 2026 and the requirement becomes national in January 2027. This evaluation covers the four years from that date.

Balance

Balanced · 0.48 previous scale

Balance on the previous scale. The Bilanz 2.0 simulation is not yet available for this evaluation. The category comes from the share of the debate on the pro side (r).

For 31 · 48 % Against 33 · 52 %

Size class: large Scale of this evaluation: Normalised Impact — unitless, calibrated to this topic. For comparison: one point here is worth roughly 2 billion euro per year. Two settings carry this balance. One death is assumed avoided for every 1,500 covered years; the published estimate for this law uses a figure three times higher, which would move the balance clearly against the measure. And the coverage the disenrolled lose is valued at 15 percent of what it costs the government, following the one study that measured what Medicaid recipients themselves would pay for it. How we score →

Arguments for

Arguments against

7 arguments evaluated · Scoring v1.3 Δ absolute −2

Arguments — For

2 arguments

Federal Medicaid spending falls

31of 100

The requirement is the single largest saving in the 2025 law's Medicaid provisions, scored at 326 billion dollars over ten years. Almost all of it comes from people leaving the programme rather than from anyone finding work. Part of it does not stay saved, because people without coverage still turn up in emergency rooms.

Value 5 · Public financesImpact 9.5Plausibility 6.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is federal money, priced at the middle of the scale like every other euro on this site. Public money is not treated as worth more or less than private money; the difference between what a euro is worth to a government and what it is worth to a household at the other end of the transfer is counted in the Impact. Nothing about what Medicaid buys enters here, because that is counted on the other side of the ledger. Nor does the deficit make the saving worth more: reducing borrowing shifts who pays and when, and that shift is inside the euro figure already. The value is the middle of the scale, the level this site uses for public money whatever it is spent on.

The budget office scores the work requirement at 326 billion dollars of reduced federal Medicaid spending over ten years, the largest single item among the law's Medicaid provisions ^[1]. That is about 32.6 billion dollars a year, or 28.1 billion euro at 1.16 dollars to the euro. The requirement begins in January 2027 and states phase it in over the following year, so about 85 percent of the annual rate is reached across the four years counted here: 23.9 billion euro. Not all of that stays saved. People who lose coverage do not stop getting ill; they arrive at emergency departments, where the bill is written off and a large share of the write-off is refunded from public funds ^[11]. A fifth is subtracted on that account, in a range from 5 to 35 percent, leaving 19.0 billion euro a year. The federal euro carries the standard weight of one. The Impact is the largest in this debate, and it is large because the saving comes from removing coverage rather than from anyone changing their behaviour.

▸ Show calculation ▾ Hide calculation

	Federal Medicaid spending the requirement removes converted at 1 euro = 1.16 dollars [1]	326 billion dollars over ten years, about 32.6 billion a year	28.1 billion euro
×	Share reached across the first four years Setting, range 70 to 100 percent: the requirement starts in January 2027 and states phase it in over the following year	85 %	23.9 billion euro
−	Care that reappears as unpaid hospital bills Setting, range 5 to 35 percent: people without coverage still arrive at emergency departments, and much of what hospitals write off is refunded from public funds [11]	4.9 billion euro, about a fifth	19 billion euro
×	Weight of a euro in the federal budget the standard weight for public money on this site, against 2.5 on the receiving side	1.0	19 billion euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	9.5

Score 9.5 Impact × 5 Value × 6.5 Plausibility ÷ 10 = 31 of 100

That fewer people on Medicaid means less Medicaid spending is arithmetic, not a prediction. What has to be shown is that people do in fact leave, and here there is more than a model. Arkansas ran the requirement for nine months in 2018: more than 18,000 adults lost coverage, and the study that measured it compared 30-to-49-year-olds, who were subject to the rule, against younger and older adults in the same state and against adults in Kentucky, Louisiana and Texas ^[3]. The confounder that would otherwise dominate — that Arkansas was on a different economic path — is what the three comparison states are there to absorb, and the age comparison inside the state absorbs the rest. Reverse causation does not arise, because the rule was imposed by the state and not chosen by the people it removed. Georgia's programme and New Hampshire's short-lived one point the same way ^[5]. What is estimated rather than observed is the size at national scale, and that doubt is carried by the range around each step rather than here. The Plausibility is high for a budget saving: the disenrolment it rests on has been observed under a comparison design, and only the national scale is projected.

evidence basis: Multiple precedents · P ceiling 8 identification: Quasi-experimental · rung ceiling 8

Counterfactual: adults not subject to the rule — younger and older age groups inside Arkansas, and adults in Kentucky, Louisiana and Texas. Design: quasi-experimental — difference-in-differences across age groups and states (Sommers and others, NEJM 2019 [3]; two-year follow-up, Health Affairs 2020 [4]). Confounder: Arkansas's own economic trajectory, absorbed by the three comparison states; age-specific trends, absorbed by the within-state comparison. Direction: no reverse causation, the rule was imposed by the state. Ceiling: quasi-experimental 8.0 binds below the several-precedents ceiling of 8.5. The size doubt sits in the phase-in and leakage bands, not in P.

More people in paid work

0.3of 100

The stated purpose of the requirement is to move people from the programme into employment. Arkansas ran the same rule for nine months and the effect was measured. It was not found.

Value 6 · OutputImpact 0.3Plausibility 2

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is output that would not otherwise exist: hours worked by people who would not have worked them. That belongs to the class this site uses for economic systems and prosperity, a step below the money it is measured in only because production is a broader good than a budget line. The earnings themselves are not the value — they are a transfer from an employer to a worker — so what is counted is the goods and services produced, net of the time the worker gives up to produce them. Whether the work is good for the person in other ways is not priced here. The value sits in the middle-upper part of the scale, at the level this site uses for economic output.

The claim is that some people subject to the requirement take a job they would otherwise not have taken. The population at risk of losing coverage is about 4.0 million on average across the four years ^[1]. If one in forty of them responds by working the required 80 hours a month rather than leaving the programme, that is 100,000 people, in a range from none to 250,000. Eighty hours a month at the wages available to this group is about 13,000 euro a year. What society gains is not the wage but the production behind it, net of the time given up, and about 40 percent is used for that share. The result is 520 million euro a year. The figure is set at the upper edge of what the Arkansas measurement could not rule out, not at the centre of what its authors found. The Impact is small even on generous assumptions, because the number of people who both would have left and would instead work is a narrow slice of a large group.

▸ Show calculation ▾ Hide calculation

	People at risk of losing coverage, yearly average [1]		4 million people
×	Share who take a job instead of leaving Setting, range 0 to 6 percent: the upper edge of what the Arkansas measurement could not rule out, not the centre of what it found [3]	2.5 %	100,000 people
×	Earnings from 80 hours a month the wages available to this group, converted at 1 euro = 1.16 dollars	13,000 euro a year	1,300 million euro
×	Production net of the time given up Setting, range 25 to 60 percent: the wage is a transfer from employer to worker, so what counts is the output behind it less the time the worker surrenders	40 %	520 million euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	0.26

Score 0.26 Impact × 6 Value × 2 Plausibility ÷ 10 = 0.3 of 100

This is the one claim in the debate that has been tested directly, and the test came back empty. Arkansas applied the same 80-hour rule to adults aged 30 to 49 in 2018 and left younger and older adults alone, which makes the comparison inside the state as well as against Kentucky, Louisiana and Texas ^[3]. Employment in the affected group did not rise relative to any of those comparisons, either in the first year or after two ^[3]^[4]. A separate analysis using national survey data reached the same conclusion. The confounder that would matter — Arkansas's own labour market moving differently — is what the comparison states control for, and reverse causation cannot arise, since nobody chose to be aged 30 to 49. The usual reply is that Arkansas ran for only nine months before a court stopped it, which is fair and is why the figure above is not set at zero. But a measurement that found nothing is not the same as no measurement, and it is what places this argument where it is. The Plausibility is low because the effect claimed here was looked for under a proper comparison and did not appear.

evidence basis: Precedent · P ceiling 8 identification: Quasi-experimental · rung ceiling 8 band: Effect did not materialise · P 1.5–2

Counterfactual: adults aged 19-29 and 50-64 inside Arkansas, plus adults in Kentucky, Louisiana and Texas. Design: quasi-experimental — difference-in-differences on age groups and states (Sommers and others, NEJM 2019 [3]; Health Affairs 2020 [4]). Confounder: Arkansas's own labour market, absorbed by the comparison states. Direction: no reverse causation, age determines exposure. Ceiling: quasi-experimental 8.0; the band binds far below it. Band: effect did not materialise — the measure's close twin ran and the employment effect did not appear.

The same rule ran in Arkansas for nine months and employment in the affected group did not rise against either comparison [3][4]. Read back: in roughly one attempt in five, an effect of the size assumed here would still show up at national scale over four years; in the rest it will not.

Open: Nebraska began in May 2026 and other states follow in 2027. An employment comparison of early against late adopters, run over two years, would settle this: a real effect would carry P to 6 or above, a second null would take the argument out of the ledger.

Arguments — Against

5 arguments · top 3 shown

Deaths among people who lose coverage

13of 100

About four million people a year lose Medicaid under the requirement, and most of them are working already but cannot document it. Coverage decides whether a cancer or a heart condition is found in time. How many deaths follow is contested by a factor of three.

Value 10 · LifeImpact 2.1Plausibility 6.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is people who die in a year and would otherwise have lived. Nothing on this site is weighted higher. The people concerned are adults between nineteen and sixty-four on very low incomes, a group with far more untreated chronic illness than the population as a whole: uncontrolled diabetes, heart disease, addiction, cancers found at a later stage. Losing coverage does not cause any of those; it decides whether the appointment that would have caught them happens. Only the deaths are priced here — illness that does not end in death is a separate argument, and the money is a third. The grief of those around each death is included at a tenth of the weight of the death itself. The value is the highest the scale allows, because the stream is human lives and nothing else is folded into it.

The budget office puts the coverage loss at 4.8 million more uninsured people from this provision alone; averaged across the four years counted here, with the phase-in, that is about 4.0 million a year ^[1]. How many deaths follow is the contested number in this debate. The measurement everyone starts from found annual mortality among low-income adults aged 55 to 64 falling 9.4 percent when Medicaid expanded ^[6]. Applied to the mortality rate of the whole 19-to-64 expansion group, which is younger but poorer and sicker than the general population, that gives one death for every 1,500 covered years — 2,667 deaths a year. The published estimate for this law is three times higher, at one death for every 460 people losing coverage, because it applies the measured effect without adjusting for the younger age mix ^[9]. A heavy age adjustment in the other direction gives one in 4,000. The range used here therefore runs from 1,000 to 5,800 deaths, and the grief of the bereaved adds a tenth on top. The Impact is the largest of the harms counted here, and its range spans a factor of six because the age adjustment is a judgment nobody has settled.

▸ Show calculation ▾ Hide calculation

	People who lose coverage, yearly average [1]	4.8 million at full effect, phased in	4 million people
÷	Deaths that follow Setting, range one per 460 to one per 4,000: annual mortality fell 9.4 percent for low-income adults aged 55 to 64 who gained Medicaid; applied to the mortality rate of the whole 19-to-64 expansion group. The published estimate for this law uses one per 460 [6][9]	one per 1,500 covered years	2,667 deaths a year
×	Value of the lives the value of a statistical life used across this site	1.4 million euro each	3,733 million euro
+	Grief of the bereaved the surcharge this site books alongside every stream of deaths	10 % of 2,667 cases × 1.4 million euro	373 million euro
=	Total per year		4,106 million euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	2.05

Score 2.05 Impact × 10 Value × 6.5 Plausibility ÷ 10 = 13 of 100

Two things have to hold: that people lose coverage, and that losing it costs lives. The first is observed. Arkansas removed more than 18,000 adults in nine months under a comparison design that held the state's own trajectory and its neighbours' constant ^[3]. The second rests on the Medicaid expansions, compared between expanding and non-expanding states before and after 2014 on death records linked to survey data, with mortality moving in parallel beforehand ^[6]. The obvious objection — that expanding states differ in ways that also move mortality — is what that parallel period tests, and it holds. Reverse causation does not arise, because expansion was a state legislative act. A randomised trial in which the tax authority wrote to uninsured households points the same way ^[7]. What is not established is the size for a population two decades younger than the one measured, and that is exactly where the published estimates and this one part company. The Plausibility is above the middle: that coverage loss costs lives is established under two independent designs, and only how many is open.

evidence basis: Converging studies · P ceiling 8 identification: Quasi-experimental · rung ceiling 8

Counterfactual: non-expansion states for the mortality link; comparison age groups and states for the disenrolment link. Design: quasi-experimental — difference-in-differences with a tested parallel pre-trend on linked survey and death-record data (Miller, Johnson and Wherry, QJE 2021 [6]), plus a difference-in-differences on the Arkansas rule itself [3]. Confounder: expansion states differ in health trends, handled by the pre-period test. Direction: no reverse causation, expansion and the work rule are both legislative acts. Ceiling: quasi-experimental 8.0 binds below the converging-studies ceiling of 9.0; the context transfer from adults aged 55-64 to the whole 19-64 group costs 1.5. The size doubt is carried by the 1,000-5,800 band, not by P.

The poorest households carry the risk again

12of 100

Everyone in the expansion group lives below 138 percent of the poverty line. Losing coverage means an illness now arrives with a bill attached, and there is nothing to pay it from. What that protection is worth is the receiving side of the saving counted on the pro side.

Value 5 · Household budgetsImpact 4.5Plausibility 5.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is money, and money is money whoever holds it, so it is priced at the middle of the scale. What is lost is a position rather than a sum: the certainty that an accident or a diagnosis will not also be a debt. That these households are the poorest in the country does not raise the value here; it raises the weight the euro carries, which is counted in the Impact. Booking their poverty in both places would price the same fact twice. The health they lose is counted in the two arguments above and is not repeated here. The value is the middle of the scale, because what is lost is a money position and the poverty of those losing it is priced in the Impact.

The federal government stops paying 23.9 billion euro a year of medical costs for these households, and the question is what share of that they lose rather than the providers who were paid it ^[1]. The one direct measurement asked what Medicaid recipients would themselves give up to keep their coverage, using the Oregon lottery: between a fifth and two fifths of what it costs, most of which is the bill rather than the protection against it ^[10]. Fifteen percent is used here as the protection share, in a range from 5 to 30 percent, giving 3.59 billion euro. Everyone in the expansion group lives below 138 percent of the poverty line, which places them in the bottom fifth of American households, where this site counts a euro at two and a half times its worth at median income. That gives 8.97 billion euro a year. The care itself, and the health it produces, is counted in the two arguments above rather than here. The Impact is second only to the saving it mirrors, and the gap between the two is the whole economic case for the measure.

▸ Show calculation ▾ Hide calculation

	Medical costs the federal government stops paying [1]	before the emergency-room leakage subtracted on the pro side	23.9 billion euro
×	Share that is protection rather than the bill itself Setting, range 5 to 30 percent: Medicaid recipients were measured to value their coverage at a fifth to two fifths of its cost, and most of that is the bill rather than the protection against it [10]	15 %	3.59 billion euro
×	Weight of a euro at these incomes everyone in the expansion group lives below 138 percent of the poverty line, in the bottom fifth of American households, where this site counts a euro at 2.5 against 1.0 in the federal budget	2.5	8.97 billion euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	4.49

Score 4.49 Impact × 5 Value × 5.5 Plausibility ÷ 10 = 12 of 100

That losing insurance exposes a household to bills it cannot pay is not in dispute; what is estimated is how much that exposure is worth. The measurement used here comes from the same Oregon draw that carries the health findings, so the counterfactual is the untreated arm of a lottery and no reverse causation or selection is possible ^[10]. Its authors separated what recipients would pay into the part that is simply the bill and the part that is protection against risk; splitting it further and applying the protection share on its own is a step this evaluation takes rather than one they take. Two things pull in opposite directions. Willingness to pay is bounded by what a household has, so for people this poor it understates what the protection is worth in any other sense. Against that, some of the care would have been written off unpaid anyway, so part of the exposure was never really theirs. The range from 5 to 30 percent is wide because both of those are unresolved. The Plausibility sits just above the middle: the underlying measurement is randomised and on this population, and the step from it to the share used here is an inference this evaluation makes on its own.

evidence basis: Study · P ceiling 8.5 identification: Experimental · rung ceiling 8.5

Counterfactual: the untreated arm of the Oregon Medicaid lottery. Design: experimental — willingness-to-pay recovered from randomised assignment (Finkelstein, Hendren and Luttmer, JPE 2019 [10]). Confounder: selection of healthier or better-off applicants into coverage, removed by the draw. Direction: no reverse causation, assignment preceded outcomes. Ceiling: experimental 8.5 binds below the studie ceiling of 9.0; a deduction of 3.0 applies because isolating a protection share and applying it alone is this evaluation's construction rather than the study's finding. The size doubt sits in the 5 to 30 percent band.

Illness that goes untreated

5.6of 100

Most of what Medicaid does never shows up in a death certificate: a depression that is treated, a prescription filled, a condition managed rather than left. The Oregon lottery measured exactly this population against a randomly chosen control group. It found a large mental-health effect and no change in blood pressure, cholesterol or blood sugar.

Value 9 · HealthImpact 1.0Plausibility 6.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is health lost short of death: depression that returns untreated, pain that goes on, a chronic condition that drifts. It belongs to the same class as the deaths above and sits one step below the top of it, because what is lost here can in principle be regained and a life cannot. What is counted is the person's own condition, not what anyone else pays as a result; the money moves in the arguments beside this one. Where an untreated depression means somebody stops working, that loss is theirs and is not booked again as an employer's. The value sits one step below the maximum: the stream is health itself, but health that can be regained.

About 4.0 million people a year lose coverage under the requirement ^[1]. What each covered year was worth in health is the only part of this that has been measured against a random control group. Oregon drew names by lot for its Medicaid expansion in 2008, and two years on the people who were drawn showed a 30 percent lower rate of depression, more preventive care and far less financial strain, with no measurable difference in blood pressure, cholesterol or blood sugar ^[8]. The depression finding alone is worth roughly 0.014 healthy life years per covered year for that group. The expansion population is somewhat less poor than Oregon's applicants, so 0.012 is used here, in a range from 0.005 to 0.020. Across 4.0 million covered years that is 48,000 healthy life years a year. Only the outcome the trial actually found is booked; the physical markers it did not find are left out. The Impact is a fifth of the deaths in weight, because a covered year buys a modest amount of measurable health across a very large number of people.

▸ Show calculation ▾ Hide calculation

	People who lose coverage, yearly average [1]		4 million people
×	Healthy life years lost Setting, range 0.005 to 0.020: the Oregon lottery measured a 30 percent fall in depression and no change in blood pressure, cholesterol or blood sugar; that finding alone is worth about 0.014 healthy life years a year for a slightly poorer group [8]	0.012 per covered year	48,000 healthy life years
×	Value of the healthy life years the value of a healthy life year used across this site	40,000 euro each	1,920 million euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	0.96

Score 0.96 Impact × 9 Value × 6.5 Plausibility ÷ 10 = 5.6 of 100

The counterfactual here is as clean as it gets in health research: the people who were not drawn in Oregon's lottery. Assignment was by lot among applicants, so the compared groups differ only in whether their name came up, and neither reverse causation nor selection into coverage can arise ^[8]. The confounder that would otherwise dominate — healthier or more organised people obtaining coverage — is removed by the draw itself. Two limits apply. The trial ran in one state, and its physical-health results were null, so a claim resting on it must stay inside what it found. This argument does, booking the mental-health stream and nothing else. The remaining stretch is that Oregon measured people gaining coverage while this argument concerns people losing it, and the two need not be symmetric — someone who has had a diagnosis and a prescription for three years loses more than a new enrollee gains. That asymmetry runs against the argument's own direction, so the figure stays where it is rather than rising. The Plausibility is above the middle: the finding is randomised and on this very population, and what is stretched is the direction rather than the setting.

evidence basis: Study · P ceiling 8.5 identification: Experimental · rung ceiling 8.5

Counterfactual: the untreated arm of the Oregon Medicaid lottery. Design: experimental — randomised assignment among applicants (Baicker and others, NEJM 2013 [8]). Confounder: healthier or better-organised applicants selecting into coverage, removed by the draw. Direction: no reverse causation, assignment preceded outcomes. Ceiling: experimental 8.5 binds below the studie ceiling of 9.0; a deduction of 2.0 applies for the direction stretch (coverage gained versus coverage lost) and for the single-state setting. The null result on physical markers is respected by booking only the depression stream.

The checking machinery costs money of its own

2of 100

Somebody has to build the systems that match employment records against enrolment files, and staff the offices that handle what the matching cannot resolve. Georgia has run this for three years: two thirds of what it spent went on administration rather than on care. The national version is cheaper per head, but it covers eighteen million people.

Value 5 · Enforcement costImpact 0.8Plausibility 5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is public money spent on running the rule rather than on anything the rule produces, priced at the middle of the scale like every other euro. It is separate from the saving on the pro side because it is a genuine cost rather than a transfer: the staff hours and the software exist and are consumed. Whether the money comes from the federal government or from the states does not change the weight, since both carry the standard weight for public money. The time the rule costs enrollees is counted separately, because it falls on different people. The value is the middle of the scale, the level this site uses for public money whatever it is spent on.

The requirement applies to roughly 18 million adults in the expansion group, each of whom must be checked at enrolment and at least twice a year afterwards ^[2]. Georgia is the only place where the full cost of doing this has been audited: through April 2025 its programme had spent 54.2 million dollars on administration against 26.1 million on health care, and less than one dollar in three of its total went to benefits ^[5]. That programme covered only a few thousand people, so its cost per head is not transferable; the building of an eligibility system is a one-off that a national scheme spreads across a far larger base. What does transfer is that the checking is not cheap. Ninety euro per person subject to the rule per year is used here, in a range from 40 to 200 — roughly what other means-tested federal programmes cost to administer per participant. Across 18 million people that is 1.62 billion euro a year. The Impact is modest against the saving but not negligible: it consumes about a twelfth of what the requirement saves.

▸ Show calculation ▾ Hide calculation

	Adults subject to the requirement [2]		18 million people
×	Cost of checking each of them Setting, range 40 to 200 euro: roughly what other means-tested federal programmes cost to administer per participant; Georgia's audited programme spent two thirds of its money on administration, but on a base of a few thousand people [5]	90 euro a year	1,620 million euro
×	Weight of a euro in public budgets the standard weight for public money on this site	1.0	1,620 million euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	0.81

Score 0.81 Impact × 5 Value × 5 Plausibility ÷ 10 = 2 of 100

That the rule costs money to run is not in question; the precedents show only that it costs a great deal more than its designers expected ^[5]. The counterfactual for Georgia is the same state's ordinary Medicaid administration, against which the Pathways programme's costs are an addition, and the audit reports both. What no source supplies is a per-person figure for a national scheme at scale, and that is what the argument needs. The 90 euro used here is built from what comparable means-tested programmes cost to administer, not measured for this one. Every link in the chain is visible — a rule requires checks, checks require data systems and staff, staff and systems cost money — and the counter-argument, that automated data matching will resolve most cases without human involvement, is real and is why the figure sits well below Georgia's. What is missing is a measurement. The Plausibility is at the top of the range for a claim whose steps are all visible and whose size nobody has counted at national scale.

evidence basis: Precedent · P ceiling 6 identification: Mechanistic · rung ceiling 6 band: Chain closed, unevidenced · P 4–5

Counterfactual: ordinary Medicaid administration without a work rule, against which Georgia's audited programme costs are an addition [5]. Design: mechanistic — the chain from rule to checks to systems and staff is named, but no source carries the national per-person figure. Confounder: Georgia's costs are dominated by a one-off eligibility-system build, handled by discarding its per-head figure and using comparable programmes instead. Direction: no reverse causation, the rule precedes the spending. Ceiling: mechanistic 6.0 binds below the praezedenz ceiling of 8.5, because the quantity has no carrier. Band: chain closed but unevidenced — the chain is closed and the automation counter-argument is answered by the low figure chosen; only the measurement is absent.

Nothing speaks against the claim, and Georgia's audit supports its direction strongly. What is missing is a per-person cost for a scheme at national scale. Read back: the checking machinery costs about what a comparable means-tested programme costs to administer, give or take a factor of two.

Open: State implementation budgets for 2027 will show the actual per-person cost within a year of the start, which would replace the setting with a measurement and could move P to 7 or above.

Reporting hours, twice a year, forever

0.9of 100

Everyone who keeps coverage under the requirement has to prove they qualify, over and over. Where employment records can be matched automatically nobody notices; where they cannot, somebody assembles pay stubs. Four million people also have to appeal a decision.

Value 9 · Life timeImpact 0.2Plausibility 5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is time people must spend and get nothing back for: finding pay stubs, uploading them, telephoning an office, sitting through an appeal. This site treats time of that kind as part of a person's life rather than as an inconvenience, which places it near the top of the scale. It is not the same as time somebody chooses to spend, and the people spending it here are among the least able to take an afternoon off. What an hour is worth and how many there are are separate questions, both set out in the derivation. Coverage lost through the same process is counted in the arguments above and not repeated here. The value is high because the stream is hours of life spent under compulsion, not a convenience that is lost.

Roughly 18 million adults are subject to the requirement and must be checked at least twice a year ^[2]. States are required to use employment and wage data they already hold before asking anyone directly, and where that works the enrollee spends no time at all; about three in five cases are assumed to clear that way, in a range from two in five to four in five. That leaves 7.2 million people assembling documents twice a year at half an hour each, or 7.2 million hours. On top of that, the four million people a year who are removed have a decision to contest, and three hours each for gathering evidence and following an appeal gives 12 million hours. Adding a further allowance for people who report but are still flagged brings the total to about 55.5 million hours a year. The hours are valued at the rate this site uses for time spent under compulsion with nothing in return. The Impact is the smallest here by a wide margin, because half an hour twice a year is a small thing spread across a very large number of people.

▸ Show calculation ▾ Hide calculation

	Adults subject to the requirement [2]		18 million people
×	Share whose employment records cannot be matched automatically Setting, range 20 to 60 percent: states must use data they already hold first, and where that works the enrollee spends no time at all	40 %	7.2 million people
×	Hours assembling documents, twice a year	2 × half an hour	7.2 million hours
+	Hours contesting a removal gathering evidence and following an appeal	4 million people × 3 hours	19.2 million hours
+	Hours of people who report and are flagged anyway Setting, range 10 to 60 million hours: the Arkansas portal was open only part of the day and reporting frequently had to be repeated [3]	allowance	55.5 million hours
×	Value of forced time the rate this site uses for time a person must spend with nothing in return	6.85 euro an hour	380 million euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	0.19

Score 0.19 Impact × 9 Value × 5 Plausibility ÷ 10 = 0.9 of 100

The chain is short and every link is visible: a documentation rule produces documentation, and documentation takes time. Arkansas supplies the counterfactual and the direction — before the rule nobody reported, after it everybody in the affected age band had to, and the state's own portal was open only part of the day, which is documented ^[3]. What no source supplies is how long it takes. Nobody has timed a Medicaid work report, and the half-hour and three-hour figures here are constructions rather than findings. The counter-argument that automated matching will make most of this invisible is a real one and is answered by assuming three in five cases clear without contact, which is the generous end of what states currently manage. Against that, the Arkansas experience suggests the people least able to report are exactly the ones the data cannot match. The Plausibility is at the top of the range for a claim whose steps are all visible and whose size nobody has measured.

evidence basis: Precedent · P ceiling 6 identification: Mechanistic · rung ceiling 6 band: Chain closed, unevidenced · P 4–5

Counterfactual: the same population before the rule, and adults outside the affected age band in Arkansas [3]. Design: mechanistic — a short chain (documentation rule → documentation → hours), with Arkansas as an observed precedent for the first two links; no source carries the hours. Confounder: automated data matching removing the burden, answered by the generous three-in-five assumption. Direction: no reverse causation, the rule precedes the reporting. Ceiling: mechanistic 6.0 binds below the praezedenz ceiling of 8.5. Band: chain closed but unevidenced — every link named, the counter-mechanism answered, only the measurement missing.

Nothing speaks against the claim; what is absent is a measurement of how long reporting takes. The one counter-mechanism, automated matching, is answered by assuming three in five cases clear without contact. Read back: reporting under this rule costs somewhere between a quarter and a full hour per episode, and roughly the assumed number of people have to do it.

Open: A time-use survey of enrollees in Nebraska, where the rule has run since May 2026, would replace both settings with measurements and could carry P to 7.

Summary

The requirement saves a great deal of money and it does so almost entirely by removing people from the programme rather than by moving anyone into work — that second effect was measured in Arkansas under a proper comparison and did not appear. So the question is whether nineteen billion euro a year of federal saving is worth what it costs the four million people who lose coverage: roughly 2,700 deaths a year, 48,000 healthy life years, and the protection that a household below the poverty line loses when an illness starts arriving with a bill. The two sides finish close, and they finish close for a specific and contestable reason: what Medicaid recipients themselves say their coverage is worth is only a fifth to two fifths of what it costs, so most of the saving does not show up as a loss to anybody in this ledger. That measure is bounded by what a very poor household is able to pay, which is not the same as what the coverage is worth to them, and the deaths carried alongside it are counted at a third of what the published estimate for this law uses. Both of those readings would have to be wrong in the same direction for the balance to come out clearly positive; either of them being right in the other direction turns it clearly negative.

Outlook — effect over time

Balanced · 0.48 previous scale

Sources

KFF: A Closer Look at the Work Requirement Provisions in the 2025 Federal Budget Reconciliation Law. kff.org
KFF: 5 Key Facts About Medicaid Work Requirements. kff.org
Sommers and others, New England Journal of Medicine 381: Medicaid Work Requirements — Results from the First Year in Arkansas. nejm.org
Sommers and others, Health Affairs 39(9): Medicaid Work Requirements In Arkansas: Two-Year Impacts On Coverage, Employment, And Affordability Of Care. healthaffairs.org
Georgia Budget and Policy Institute: Pathways to Coverage: Looking Back Two Years and Into the Future. gbpi.org
Miller, Johnson and Wherry, Quarterly Journal of Economics 136(3): Medicaid and Mortality: New Evidence From Linked Survey and Administrative Data. academic.oup.com
Goldin, Lurie and McCubbin, Quarterly Journal of Economics 136(1): Health Insurance and Mortality: Experimental Evidence from Taxpayer Outreach. academic.oup.com
Baicker and others, New England Journal of Medicine 368: The Oregon Experiment: Effects of Medicaid on Clinical Outcomes. nejm.org
Gaffney, Himmelstein and Woolhandler, Annals of Internal Medicine: Projected Effects of Proposed Cuts in Federal Medicaid Expenditures on Medicaid Enrollment, Uninsurance, Health Care, and Health. acpjournals.org
Finkelstein, Hendren and Luttmer, Journal of Political Economy 127(6): The Value of Medicaid: Interpreting Results from the Oregon Health Insurance Experiment. journals.uchicago.edu
Commonwealth Fund: The Impact of Proposed Federal Medicaid Work Requirements on Hospital Revenues and Financial Margins. commonwealthfund.org

Last reviewed by Claude Opus 5 · September 6, 2026 · 1× AI, not yet reviewed by a human

September 6, 2026AI reviewClaude Opus 5First evaluation
Created for the English side: the employment claim placed on the finding ladder against the Arkansas measurement.

Evaluations are produced with AI support and reviewed on a schedule for new developments; human passes are marked separately.How we review →